Speech recognition method and apparatus

Abstract

A speech recognition method includes receiving speech data, obtaining, from the received speech data, a candidate text including at least one word and a phonetic symbol sequence associated with a pronunciation of a target word included in the received speech data, using a speech recognition model, replacing the phonetic symbol sequence included in the candidate text with a replacement word corresponding to the phonetic symbol sequence, and determining a target text corresponding to the received speech data based on a result of the replacing.

What is claimed is: 
     
       1. A speech recognition method comprising:
 obtaining a speech recognition model by:
 detecting proper nouns from training data, 
 replacing, using dictionary data corresponding to a category indicated by an identifier pair among sets of dictionary data corresponding to different categories, each proper noun with a phonetic symbol sequence associated with each proper noun to obtain preprocessed training data, and 
 training the speech recognition model based on the preprocessed training data; 
 
 receiving speech data; 
 obtaining, from the received speech data and using the speech recognition model, a candidate text including at least one word, a phonetic symbol sequence associated with a pronunciation of a target word included in the received speech data, and the identifier pair, wherein the identifier pair indicates a start and an end of the phonetic symbol sequence, and the category of the target word; 
 determining a replacement word corresponding to the phonetic symbol sequence included in the candidate text, using the dictionary data corresponding to the category indicated by the identifier pair among the sets of dictionary data corresponding to the different categories; 
 replacing the phonetic symbol sequence with the replacement word; and 
 determining a target text corresponding to the received speech data based on a result of the replacing, 
 wherein each of the sets of dictionary data corresponding to the different categories comprises information associated with a phonetic symbol sequence corresponding to each of words in each of the categories. 
 
     
     
       2. The speech recognition method of  claim 1 , wherein the at least one word comprises at least one subword, and
 the candidate text comprises the at least one subword, and the phonetic symbol sequence. 
 
     
     
       3. The speech recognition method of  claim 2 , wherein the replacing comprises:
 replacing, with the replacement word, the phonetic symbol sequence identified by the identifier pair. 
 
     
     
       4. The speech recognition method of  claim 1 , further comprising:
 determining the replacement word corresponding to the phonetic symbol sequence using dictionary data including information associated with a plurality of words and phonetic symbol sequences respectively corresponding to the words. 
 
     
     
       5. The speech recognition method of  claim 4 , further comprising:
 calculating a similarity between the phonetic symbol sequence included in the candidate text and each of the phonetic symbol sequences included in the dictionary data; and 
 determining, as the replacement word corresponding to the phonetic symbol sequence included in the candidate text, a word corresponding to a phonetic symbol sequence having a greatest similarity among calculated similarities of the phonetic symbol sequences included in the dictionary data. 
 
     
     
       6. The speech recognition method of  claim 4 , wherein the dictionary data is of a trie or hashmap data structure, and
 the determining comprises:
 retrieving a phonetic symbol sequence corresponding to the phonetic symbol sequence included in the candidate text from the phonetic symbol sequences included in the dictionary data, using the data structure; and 
 determining a word corresponding to the retrieved phonetic symbol sequence to be the replacement word corresponding to the phonetic symbol sequence included in the candidate text. 
 
 
     
     
       7. The speech recognition method of  claim 1 , further comprising:
 obtaining a plurality of candidate target texts by replacing the phonetic symbol sequence with each of the words, in response to the phonetic symbol sequence corresponding to a plurality of words; 
 calculating a score of each of the candidate target texts using a language model; and 
 determining, to be the target text, a candidate target text having a greatest score among calculated scores of the candidate target texts. 
 
     
     
       8. The speech recognition method of  claim 1 , wherein the phonetic symbol sequence is associated with a pronunciation of the target word corresponding to a proper noun. 
     
     
       9. The speech recognition method of  claim 1 , wherein the speech recognition model comprises:
 an encoder configured to extract a vector value from the received speech data; and 
 a decoder configured to output the candidate text corresponding to the received speech data based on the vector value. 
 
     
     
       10. A non-transitory computer-readable storage medium storing instructions that, when executed by a processor, cause the processor to perform the speech recognition method of  claim 1 . 
     
     
       11. A speech recognition apparatus comprising: a processor configured to:
 obtain a speech recognition model by:
 detecting proper nouns from training data, 
 replacing, using dictionary data corresponding to a category indicated by an identifier pair among sets of dictionary data corresponding to different categories, each proper noun with a phonetic symbol sequence associated with each proper noun to obtain preprocessed training data, and 
 training the speech recognition model based on the preprocessed training data; 
 
 receive speech data; 
 obtain, from the received speech data and using the speech recognition model, a candidate text including at least one word, a phonetic symbol sequence associated with a pronunciation of a target word included in the received speech data, and the identifier pair, wherein the identifier pair indicates a start and an end of the phonetic symbol sequence and the category of the target word; 
 determine a replacement word corresponding to the phonetic symbol sequence using the dictionary data corresponding to the category indicated by the identifier pair among the sets of dictionary data respectively corresponding to the different categories; 
 replace the phonetic symbol sequence with the replacement word; and 
 determine a target text corresponding to the received speech data based on a result of the replacing, 
 wherein each of the sets of dictionary data corresponding to the different categories comprises information associated with a phonetic symbol sequence corresponding to each of words in each of the categories. 
 
     
     
       12. The speech recognition apparatus of  claim 11 , wherein the at least one word comprises at least one subword, and
 the candidate text comprises the at least one subword, the phonetic symbol sequence. 
 
     
     
       13. The speech recognition apparatus of  claim 12 , wherein the processor is further configured to:
 replace, with the replacement word, the phonetic symbol sequence identified by the identifier pair. 
 
     
     
       14. The speech recognition apparatus of  claim 11 , wherein the processor is further configured to:
 determine the replacement word corresponding to the phonetic symbol sequence, using dictionary data including information associated with a plurality of words and a phonetic symbol sequence corresponding to each of the words. 
 
     
     
       15. The speech recognition apparatus of  claim 11 , wherein the processor is further configured to:
 obtain a plurality of candidate target texts by replacing the phonetic symbol sequence with each of the words, in response to the phonetic symbol sequence corresponding to a plurality of words; 
 calculate a score of each of the candidate target texts using a language model; and 
 determine, to be the target text, a candidate target text having a greatest score among calculated scores of the candidate target texts. 
 
     
     
       16. A speech recognition method comprising:
 obtaining a speech recognition model by:
 detecting proper nouns from training data, 
 replacing, using dictionary data corresponding to a category indicated by an identifier pair among sets of dictionary data corresponding to different categories, each proper noun with a phonetic symbol sequence associated with each proper noun to obtain preprocessed training data, and 
 training the speech recognition model based on the preprocessed training data; 
 
 receiving speech data; 
 obtaining, from the received speech data and using the speech recognition model, a target phonetic symbol sequence that represents a pronunciation of a target word included in the received speech data and the identifier pair indicates a start and an end of the target phonetic symbol sequence and the category of the target word; 
 determining a replacement word corresponding to the phonetic symbol sequence included in the candidate text, using the dictionary data corresponding to the category indicated by the identifier pair among the sets of dictionary data corresponding to the different categories; and 
 outputting a target text corresponding to the received speech data by replacing the target word with the replacement words, based on the determining, 
 wherein each of the sets of dictionary data corresponding to the different categories comprises information associated with a phonetic symbol sequence corresponding to each of words in each of the categories. 
 
     
     
       17. The speech recognition method of  claim 16 , further comprising determining the replacement word to replace the target word as one of the respective replacements words that is associated with one of the other phonetic symbol sequences that is most similar to the target phonetic symbol sequence. 
     
     
       18. The speech recognition method of  claim 16 , wherein the target phonetic symbol sequence is included in candidate text comprising at least one subword that precedes or follows the target phonetic symbol sequence. 
     
     
       19. The speech recognition method of  claim 18 , wherein the identifier pair comprises a first character that indicates a start of the target phonetic symbol sequence and a second character that that indicates an end of the target phonetic symbol sequence, and
 the at least one subword is separated from the target phonetic symbol sequence in the candidate text by one or both of the first character and the second character.